Every-day decisions frequently require choosing among multiple alternatives. Compared to binary choice paradigms, much less is known about the computational principle of decisions with more than two options. Previous physiological and behavioral experiments have revealed puzzling properties of human/animal decisions involving more than two options, such as interactions among these options and time-dependent decision thresholds. Why the nervous system ought to have such properties and how they functionally relate to each other remains poorly understood. To address these problems, we have derived the normative strategies for general value-based decisions with multiple options, and have identified the optimal stopping-rules for value-based evidence accumulation. The resulting strategies appear complex but turn out to be well approximated by a remarkably simple neural mechanism, with time-dependent activity-normalization in a recurrent circuit controlled by an urgency signal. Our model thus reveals why the nervous system requires such activity normalization and urgency signal: they allow the nervous system to implement efficient decisions under multi-alternative choices. The model predicts a time-dependent normalization that constrains neural population activity during decision-making.